基于多特征融合和小波去噪的镜头渐变检测方法  

Approach of gradual shot-boundary based on multi-feature fusion and wavelet denoising

作　　者：何利[1] 刘群[1] 

机构地区：[1]重庆邮电大学计算机科学与技术学院,重庆400065

出　　处：《重庆邮电大学学报（自然科学版）》2009年第5期690-694,共5页Journal of Chongqing University of Posts and Telecommunications(Natural Science Edition)

基　　金：重庆市自然科学基重庆市自然科学基金(CSTC2007BB2386);重庆市教委应用基础研究项目(KJ070504KJ080510);重庆邮电大学青年自然科学基金项目(A2008-34)

摘　　要：镜头检测在视频检索中起着关键的作用。镜头渐变检测已有许多成熟的方法,如何识别镜头渐变是视频分割中的重点。将视频的多帧差序列视为一维信号,提出了一种基于多特征融合和小波去噪的二次帧差滑动窗口镜头边界检测算法,该方法能同时检测出镜头的切变和渐变,实验表明该方法在一定程度上克服了由于摄像机不规则运动、光影变化、画面中物体快速运动等造成的影响,通过对不同视频片段的测试和比较,表明该算法能较好地实现镜头的突变和渐变检测。Shot detection plays an important role in video retrieval. Some mature cut shot detection methods have been presented, and how to identify the gradual change in video sequences is the key point. Treating the multi-fame difference sequence of video as one dimensional signal, a shot boundary detection algorithm of twice difference sliding window based on multi-feature fusion and wavelet denoising was proposed. The new approach can detect the cut and gradual change simultaneously. And experiments show that this approach overcomes the influence caused by irregular motion of camera, change of light and shadow, and fast movement in pictures. Through the test and comparison of different video clips, it is shown that this algorithm can realize the detection on abrupt and gradual shot change.

关 键 词：镜头渐变检测 特征融合 小波去噪 二次帧差 

分 类 号：TP37[自动化与计算机技术—计算机系统结构]